We propose the BeleniosCaI protocol, a variant of Belenios which brings the cast-as-intended property, in addition to other existing security properties. Our approach is based on a 2-part checksum that the voting device commits to, before being challenged to reveal one of them chosen at random by the voter. It requires only one device on the voter's side and does not rely on previously sent data like with return codes. Compared to the classical Benaloh auditor cast approach, we still have cast-as-intended with only some probability, but the voter's journey is more linear, and the audited ballot is really the one that is cast. We formally prove the security of BeleniosCaI w.r.t. end-to-end verifiability and privacy in a symbolic model, using the ProVerif tool.